Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Curiosity-driven Exploration by Self-supervised Prediction
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use ICM curiosity with A3C instead of A3C alone | A3C · VizDoom sparse reward | Mean score: 0.0 % -> 100.0 % | Helped |
| Add VIME exploration to TRPO | TRPO · VizDoom sparse reward | Mean score: 26.0 % -> 46.1 % | Helped |
| Use feature-space curiosity rather than pixel-space prediction error (ablation) | ICM + A3C · VizDoom very sparse reward | Success: 66% | Helped |
| Remove the inverse model and predict the next observation in pixel space (ablation) | ICM + A3C · VizDoom very sparse reward | Success | Hurt |
| Add a region of white-noise distractor to observations | ICM-pixels + A3C · VizDoom sparse reward with distractor | Score: 40% | Hurt |
| Pretrain on Mario Level-1, then fine-tune on Level-2 with curiosity | Curiosity-driven Mario policy · Mario Level-2 | Mean distance: 399.7 ± 22.5 -> 466 ± 37.9 | Helped |
| Pretrain on Mario Level-1, then fine-tune on Level-3 with curiosity | Curiosity-driven Mario policy · Mario Level-3 | Mean distance: 319.3 ± 9.7 -> 97.5 ± 17.4 | Hurt |
| Pretrain with curiosity on a separate VizDoom map, then fine-tune with extrinsic reward | ICM + A3C · VizDoom very sparse reward on novel map | Reward | Helped |
| Replace the learned curiosity reward with random noise (ablation) | A3C curiosity agent · VizDoom sparse reward | Goal success | Hurt |
Closest papers by meaning
- Large-Scale Study of Curiosity-Driven Learning (ICLR 2019 · similarity 0.91)
- Wonder Wins Ways: Curiosity-Driven Exploration through Multi-Agent Contextual Calibration (NeurIPS 2025 · similarity 0.90)
- DISCOVER: Automated Curricula for Sparse-Reward Reinforcement Learning (NeurIPS 2025 · similarity 0.88)
- Efficient Exploration and Discriminative World Model Learning with an Object-Centric Abstraction (ICLR 2025 · similarity 0.87)
- Behavioral Exploration: Learning to Explore via In-Context Adaptation (ICML 2025 · similarity 0.87)
- A Single Goal is All You Need: Skills and Exploration Emerge from Contrastive RL without Rewards, Demonstrations, or Subgoals (ICLR 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.